Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | |
6728cb0e JA |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
7 | * - July2000 | |
1da177e4 LT |
8 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
9 | */ | |
10 | ||
11 | /* | |
12 | * This handles all read/write requests to block devices | |
13 | */ | |
1da177e4 LT |
14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/backing-dev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/blkdev.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/kernel_stat.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/slab.h> | |
26 | #include <linux/swap.h> | |
27 | #include <linux/writeback.h> | |
faccbd4b | 28 | #include <linux/task_io_accounting_ops.h> |
2056a782 | 29 | #include <linux/blktrace_api.h> |
c17bb495 | 30 | #include <linux/fault-inject.h> |
5f3ea37c | 31 | #include <trace/block.h> |
1da177e4 | 32 | |
8324aa91 JA |
33 | #include "blk.h" |
34 | ||
0bfc2455 IM |
35 | DEFINE_TRACE(block_plug); |
36 | DEFINE_TRACE(block_unplug_io); | |
37 | DEFINE_TRACE(block_unplug_timer); | |
38 | DEFINE_TRACE(block_getrq); | |
39 | DEFINE_TRACE(block_sleeprq); | |
40 | DEFINE_TRACE(block_rq_requeue); | |
41 | DEFINE_TRACE(block_bio_backmerge); | |
42 | DEFINE_TRACE(block_bio_frontmerge); | |
43 | DEFINE_TRACE(block_bio_queue); | |
44 | DEFINE_TRACE(block_rq_complete); | |
45 | DEFINE_TRACE(block_remap); /* Also used in drivers/md/dm.c */ | |
46 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_remap); | |
47 | ||
165125e1 | 48 | static int __make_request(struct request_queue *q, struct bio *bio); |
1da177e4 LT |
49 | |
50 | /* | |
51 | * For the allocated request tables | |
52 | */ | |
5ece6c52 | 53 | static struct kmem_cache *request_cachep; |
1da177e4 LT |
54 | |
55 | /* | |
56 | * For queue allocation | |
57 | */ | |
6728cb0e | 58 | struct kmem_cache *blk_requestq_cachep; |
1da177e4 | 59 | |
1da177e4 LT |
60 | /* |
61 | * Controlling structure to kblockd | |
62 | */ | |
ff856bad | 63 | static struct workqueue_struct *kblockd_workqueue; |
1da177e4 | 64 | |
26b8256e JA |
65 | static void drive_stat_acct(struct request *rq, int new_io) |
66 | { | |
28f13702 | 67 | struct hd_struct *part; |
26b8256e | 68 | int rw = rq_data_dir(rq); |
c9959059 | 69 | int cpu; |
26b8256e | 70 | |
26308eab | 71 | if (!blk_fs_request(rq) || !blk_do_io_stat(rq)) |
26b8256e JA |
72 | return; |
73 | ||
074a7aca | 74 | cpu = part_stat_lock(); |
e71bf0d0 | 75 | part = disk_map_sector_rcu(rq->rq_disk, rq->sector); |
c9959059 | 76 | |
28f13702 | 77 | if (!new_io) |
074a7aca | 78 | part_stat_inc(cpu, part, merges[rw]); |
28f13702 | 79 | else { |
074a7aca TH |
80 | part_round_stats(cpu, part); |
81 | part_inc_in_flight(part); | |
26b8256e | 82 | } |
e71bf0d0 | 83 | |
074a7aca | 84 | part_stat_unlock(); |
26b8256e JA |
85 | } |
86 | ||
8324aa91 | 87 | void blk_queue_congestion_threshold(struct request_queue *q) |
1da177e4 LT |
88 | { |
89 | int nr; | |
90 | ||
91 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | |
92 | if (nr > q->nr_requests) | |
93 | nr = q->nr_requests; | |
94 | q->nr_congestion_on = nr; | |
95 | ||
96 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | |
97 | if (nr < 1) | |
98 | nr = 1; | |
99 | q->nr_congestion_off = nr; | |
100 | } | |
101 | ||
1da177e4 LT |
102 | /** |
103 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | |
104 | * @bdev: device | |
105 | * | |
106 | * Locates the passed device's request queue and returns the address of its | |
107 | * backing_dev_info | |
108 | * | |
109 | * Will return NULL if the request queue cannot be located. | |
110 | */ | |
111 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | |
112 | { | |
113 | struct backing_dev_info *ret = NULL; | |
165125e1 | 114 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
115 | |
116 | if (q) | |
117 | ret = &q->backing_dev_info; | |
118 | return ret; | |
119 | } | |
1da177e4 LT |
120 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
121 | ||
2a4aa30c | 122 | void blk_rq_init(struct request_queue *q, struct request *rq) |
1da177e4 | 123 | { |
1afb20f3 FT |
124 | memset(rq, 0, sizeof(*rq)); |
125 | ||
1da177e4 | 126 | INIT_LIST_HEAD(&rq->queuelist); |
242f9dcb | 127 | INIT_LIST_HEAD(&rq->timeout_list); |
c7c22e4d | 128 | rq->cpu = -1; |
63a71386 JA |
129 | rq->q = q; |
130 | rq->sector = rq->hard_sector = (sector_t) -1; | |
2e662b65 JA |
131 | INIT_HLIST_NODE(&rq->hash); |
132 | RB_CLEAR_NODE(&rq->rb_node); | |
d7e3c324 | 133 | rq->cmd = rq->__cmd; |
e2494e1b | 134 | rq->cmd_len = BLK_MAX_CDB; |
63a71386 | 135 | rq->tag = -1; |
1da177e4 | 136 | rq->ref_count = 1; |
1da177e4 | 137 | } |
2a4aa30c | 138 | EXPORT_SYMBOL(blk_rq_init); |
1da177e4 | 139 | |
5bb23a68 N |
140 | static void req_bio_endio(struct request *rq, struct bio *bio, |
141 | unsigned int nbytes, int error) | |
1da177e4 | 142 | { |
165125e1 | 143 | struct request_queue *q = rq->q; |
797e7dbb | 144 | |
5bb23a68 N |
145 | if (&q->bar_rq != rq) { |
146 | if (error) | |
147 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
148 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
149 | error = -EIO; | |
797e7dbb | 150 | |
5bb23a68 | 151 | if (unlikely(nbytes > bio->bi_size)) { |
6728cb0e | 152 | printk(KERN_ERR "%s: want %u bytes done, %u left\n", |
24c03d47 | 153 | __func__, nbytes, bio->bi_size); |
5bb23a68 N |
154 | nbytes = bio->bi_size; |
155 | } | |
797e7dbb | 156 | |
08bafc03 KM |
157 | if (unlikely(rq->cmd_flags & REQ_QUIET)) |
158 | set_bit(BIO_QUIET, &bio->bi_flags); | |
159 | ||
5bb23a68 N |
160 | bio->bi_size -= nbytes; |
161 | bio->bi_sector += (nbytes >> 9); | |
7ba1ba12 MP |
162 | |
163 | if (bio_integrity(bio)) | |
164 | bio_integrity_advance(bio, nbytes); | |
165 | ||
5bb23a68 | 166 | if (bio->bi_size == 0) |
6712ecf8 | 167 | bio_endio(bio, error); |
5bb23a68 N |
168 | } else { |
169 | ||
170 | /* | |
171 | * Okay, this is the barrier request in progress, just | |
172 | * record the error; | |
173 | */ | |
174 | if (error && !q->orderr) | |
175 | q->orderr = error; | |
176 | } | |
1da177e4 | 177 | } |
1da177e4 | 178 | |
1da177e4 LT |
179 | void blk_dump_rq_flags(struct request *rq, char *msg) |
180 | { | |
181 | int bit; | |
182 | ||
6728cb0e | 183 | printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, |
4aff5e23 JA |
184 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, |
185 | rq->cmd_flags); | |
1da177e4 | 186 | |
6728cb0e JA |
187 | printk(KERN_INFO " sector %llu, nr/cnr %lu/%u\n", |
188 | (unsigned long long)rq->sector, | |
189 | rq->nr_sectors, | |
190 | rq->current_nr_sectors); | |
191 | printk(KERN_INFO " bio %p, biotail %p, buffer %p, data %p, len %u\n", | |
192 | rq->bio, rq->biotail, | |
193 | rq->buffer, rq->data, | |
194 | rq->data_len); | |
1da177e4 | 195 | |
4aff5e23 | 196 | if (blk_pc_request(rq)) { |
6728cb0e | 197 | printk(KERN_INFO " cdb: "); |
d34c87e4 | 198 | for (bit = 0; bit < BLK_MAX_CDB; bit++) |
1da177e4 LT |
199 | printk("%02x ", rq->cmd[bit]); |
200 | printk("\n"); | |
201 | } | |
202 | } | |
1da177e4 LT |
203 | EXPORT_SYMBOL(blk_dump_rq_flags); |
204 | ||
1da177e4 LT |
205 | /* |
206 | * "plug" the device if there are no outstanding requests: this will | |
207 | * force the transfer to start only after we have put all the requests | |
208 | * on the list. | |
209 | * | |
210 | * This is called with interrupts off and no requests on the queue and | |
211 | * with the queue lock held. | |
212 | */ | |
165125e1 | 213 | void blk_plug_device(struct request_queue *q) |
1da177e4 LT |
214 | { |
215 | WARN_ON(!irqs_disabled()); | |
216 | ||
217 | /* | |
218 | * don't plug a stopped queue, it must be paired with blk_start_queue() | |
219 | * which will restart the queueing | |
220 | */ | |
7daac490 | 221 | if (blk_queue_stopped(q)) |
1da177e4 LT |
222 | return; |
223 | ||
e48ec690 | 224 | if (!queue_flag_test_and_set(QUEUE_FLAG_PLUGGED, q)) { |
1da177e4 | 225 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
5f3ea37c | 226 | trace_block_plug(q); |
2056a782 | 227 | } |
1da177e4 | 228 | } |
1da177e4 LT |
229 | EXPORT_SYMBOL(blk_plug_device); |
230 | ||
6c5e0c4d JA |
231 | /** |
232 | * blk_plug_device_unlocked - plug a device without queue lock held | |
233 | * @q: The &struct request_queue to plug | |
234 | * | |
235 | * Description: | |
236 | * Like @blk_plug_device(), but grabs the queue lock and disables | |
237 | * interrupts. | |
238 | **/ | |
239 | void blk_plug_device_unlocked(struct request_queue *q) | |
240 | { | |
241 | unsigned long flags; | |
242 | ||
243 | spin_lock_irqsave(q->queue_lock, flags); | |
244 | blk_plug_device(q); | |
245 | spin_unlock_irqrestore(q->queue_lock, flags); | |
246 | } | |
247 | EXPORT_SYMBOL(blk_plug_device_unlocked); | |
248 | ||
1da177e4 LT |
249 | /* |
250 | * remove the queue from the plugged list, if present. called with | |
251 | * queue lock held and interrupts disabled. | |
252 | */ | |
165125e1 | 253 | int blk_remove_plug(struct request_queue *q) |
1da177e4 LT |
254 | { |
255 | WARN_ON(!irqs_disabled()); | |
256 | ||
e48ec690 | 257 | if (!queue_flag_test_and_clear(QUEUE_FLAG_PLUGGED, q)) |
1da177e4 LT |
258 | return 0; |
259 | ||
260 | del_timer(&q->unplug_timer); | |
261 | return 1; | |
262 | } | |
1da177e4 LT |
263 | EXPORT_SYMBOL(blk_remove_plug); |
264 | ||
265 | /* | |
266 | * remove the plug and let it rip.. | |
267 | */ | |
165125e1 | 268 | void __generic_unplug_device(struct request_queue *q) |
1da177e4 | 269 | { |
7daac490 | 270 | if (unlikely(blk_queue_stopped(q))) |
1da177e4 | 271 | return; |
a31a9738 | 272 | if (!blk_remove_plug(q) && !blk_queue_nonrot(q)) |
1da177e4 LT |
273 | return; |
274 | ||
22e2c507 | 275 | q->request_fn(q); |
1da177e4 | 276 | } |
1da177e4 LT |
277 | |
278 | /** | |
279 | * generic_unplug_device - fire a request queue | |
165125e1 | 280 | * @q: The &struct request_queue in question |
1da177e4 LT |
281 | * |
282 | * Description: | |
283 | * Linux uses plugging to build bigger requests queues before letting | |
284 | * the device have at them. If a queue is plugged, the I/O scheduler | |
285 | * is still adding and merging requests on the queue. Once the queue | |
286 | * gets unplugged, the request_fn defined for the queue is invoked and | |
287 | * transfers started. | |
288 | **/ | |
165125e1 | 289 | void generic_unplug_device(struct request_queue *q) |
1da177e4 | 290 | { |
dbaf2c00 JA |
291 | if (blk_queue_plugged(q)) { |
292 | spin_lock_irq(q->queue_lock); | |
293 | __generic_unplug_device(q); | |
294 | spin_unlock_irq(q->queue_lock); | |
295 | } | |
1da177e4 LT |
296 | } |
297 | EXPORT_SYMBOL(generic_unplug_device); | |
298 | ||
299 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | |
300 | struct page *page) | |
301 | { | |
165125e1 | 302 | struct request_queue *q = bdi->unplug_io_data; |
1da177e4 | 303 | |
2ad8b1ef | 304 | blk_unplug(q); |
1da177e4 LT |
305 | } |
306 | ||
86db1e29 | 307 | void blk_unplug_work(struct work_struct *work) |
1da177e4 | 308 | { |
165125e1 JA |
309 | struct request_queue *q = |
310 | container_of(work, struct request_queue, unplug_work); | |
1da177e4 | 311 | |
5f3ea37c | 312 | trace_block_unplug_io(q); |
1da177e4 LT |
313 | q->unplug_fn(q); |
314 | } | |
315 | ||
86db1e29 | 316 | void blk_unplug_timeout(unsigned long data) |
1da177e4 | 317 | { |
165125e1 | 318 | struct request_queue *q = (struct request_queue *)data; |
1da177e4 | 319 | |
5f3ea37c | 320 | trace_block_unplug_timer(q); |
18887ad9 | 321 | kblockd_schedule_work(q, &q->unplug_work); |
1da177e4 LT |
322 | } |
323 | ||
2ad8b1ef AB |
324 | void blk_unplug(struct request_queue *q) |
325 | { | |
326 | /* | |
327 | * devices don't necessarily have an ->unplug_fn defined | |
328 | */ | |
329 | if (q->unplug_fn) { | |
5f3ea37c | 330 | trace_block_unplug_io(q); |
2ad8b1ef AB |
331 | q->unplug_fn(q); |
332 | } | |
333 | } | |
334 | EXPORT_SYMBOL(blk_unplug); | |
335 | ||
1da177e4 LT |
336 | /** |
337 | * blk_start_queue - restart a previously stopped queue | |
165125e1 | 338 | * @q: The &struct request_queue in question |
1da177e4 LT |
339 | * |
340 | * Description: | |
341 | * blk_start_queue() will clear the stop flag on the queue, and call | |
342 | * the request_fn for the queue if it was in a stopped state when | |
343 | * entered. Also see blk_stop_queue(). Queue lock must be held. | |
344 | **/ | |
165125e1 | 345 | void blk_start_queue(struct request_queue *q) |
1da177e4 | 346 | { |
a038e253 PBG |
347 | WARN_ON(!irqs_disabled()); |
348 | ||
75ad23bc | 349 | queue_flag_clear(QUEUE_FLAG_STOPPED, q); |
a538cd03 | 350 | __blk_run_queue(q); |
1da177e4 | 351 | } |
1da177e4 LT |
352 | EXPORT_SYMBOL(blk_start_queue); |
353 | ||
354 | /** | |
355 | * blk_stop_queue - stop a queue | |
165125e1 | 356 | * @q: The &struct request_queue in question |
1da177e4 LT |
357 | * |
358 | * Description: | |
359 | * The Linux block layer assumes that a block driver will consume all | |
360 | * entries on the request queue when the request_fn strategy is called. | |
361 | * Often this will not happen, because of hardware limitations (queue | |
362 | * depth settings). If a device driver gets a 'queue full' response, | |
363 | * or if it simply chooses not to queue more I/O at one point, it can | |
364 | * call this function to prevent the request_fn from being called until | |
365 | * the driver has signalled it's ready to go again. This happens by calling | |
366 | * blk_start_queue() to restart queue operations. Queue lock must be held. | |
367 | **/ | |
165125e1 | 368 | void blk_stop_queue(struct request_queue *q) |
1da177e4 LT |
369 | { |
370 | blk_remove_plug(q); | |
75ad23bc | 371 | queue_flag_set(QUEUE_FLAG_STOPPED, q); |
1da177e4 LT |
372 | } |
373 | EXPORT_SYMBOL(blk_stop_queue); | |
374 | ||
375 | /** | |
376 | * blk_sync_queue - cancel any pending callbacks on a queue | |
377 | * @q: the queue | |
378 | * | |
379 | * Description: | |
380 | * The block layer may perform asynchronous callback activity | |
381 | * on a queue, such as calling the unplug function after a timeout. | |
382 | * A block device may call blk_sync_queue to ensure that any | |
383 | * such activity is cancelled, thus allowing it to release resources | |
59c51591 | 384 | * that the callbacks might use. The caller must already have made sure |
1da177e4 LT |
385 | * that its ->make_request_fn will not re-add plugging prior to calling |
386 | * this function. | |
387 | * | |
388 | */ | |
389 | void blk_sync_queue(struct request_queue *q) | |
390 | { | |
391 | del_timer_sync(&q->unplug_timer); | |
70ed28b9 | 392 | del_timer_sync(&q->timeout); |
64d01dc9 | 393 | cancel_work_sync(&q->unplug_work); |
1da177e4 LT |
394 | } |
395 | EXPORT_SYMBOL(blk_sync_queue); | |
396 | ||
397 | /** | |
80a4b58e | 398 | * __blk_run_queue - run a single device queue |
1da177e4 | 399 | * @q: The queue to run |
80a4b58e JA |
400 | * |
401 | * Description: | |
402 | * See @blk_run_queue. This variant must be called with the queue lock | |
403 | * held and interrupts disabled. | |
404 | * | |
1da177e4 | 405 | */ |
75ad23bc | 406 | void __blk_run_queue(struct request_queue *q) |
1da177e4 | 407 | { |
1da177e4 | 408 | blk_remove_plug(q); |
dac07ec1 | 409 | |
a538cd03 TH |
410 | if (unlikely(blk_queue_stopped(q))) |
411 | return; | |
412 | ||
413 | if (elv_queue_empty(q)) | |
414 | return; | |
415 | ||
dac07ec1 JA |
416 | /* |
417 | * Only recurse once to avoid overrunning the stack, let the unplug | |
418 | * handling reinvoke the handler shortly if we already got there. | |
419 | */ | |
a538cd03 TH |
420 | if (!queue_flag_test_and_set(QUEUE_FLAG_REENTER, q)) { |
421 | q->request_fn(q); | |
422 | queue_flag_clear(QUEUE_FLAG_REENTER, q); | |
423 | } else { | |
424 | queue_flag_set(QUEUE_FLAG_PLUGGED, q); | |
425 | kblockd_schedule_work(q, &q->unplug_work); | |
426 | } | |
75ad23bc NP |
427 | } |
428 | EXPORT_SYMBOL(__blk_run_queue); | |
dac07ec1 | 429 | |
75ad23bc NP |
430 | /** |
431 | * blk_run_queue - run a single device queue | |
432 | * @q: The queue to run | |
80a4b58e JA |
433 | * |
434 | * Description: | |
435 | * Invoke request handling on this queue, if it has pending work to do. | |
a7f55792 | 436 | * May be used to restart queueing when a request has completed. |
75ad23bc NP |
437 | */ |
438 | void blk_run_queue(struct request_queue *q) | |
439 | { | |
440 | unsigned long flags; | |
441 | ||
442 | spin_lock_irqsave(q->queue_lock, flags); | |
443 | __blk_run_queue(q); | |
1da177e4 LT |
444 | spin_unlock_irqrestore(q->queue_lock, flags); |
445 | } | |
446 | EXPORT_SYMBOL(blk_run_queue); | |
447 | ||
165125e1 | 448 | void blk_put_queue(struct request_queue *q) |
483f4afc AV |
449 | { |
450 | kobject_put(&q->kobj); | |
451 | } | |
483f4afc | 452 | |
6728cb0e | 453 | void blk_cleanup_queue(struct request_queue *q) |
483f4afc | 454 | { |
e3335de9 JA |
455 | /* |
456 | * We know we have process context here, so we can be a little | |
457 | * cautious and ensure that pending block actions on this device | |
458 | * are done before moving on. Going into this function, we should | |
459 | * not have processes doing IO to this device. | |
460 | */ | |
461 | blk_sync_queue(q); | |
462 | ||
483f4afc | 463 | mutex_lock(&q->sysfs_lock); |
75ad23bc | 464 | queue_flag_set_unlocked(QUEUE_FLAG_DEAD, q); |
483f4afc AV |
465 | mutex_unlock(&q->sysfs_lock); |
466 | ||
467 | if (q->elevator) | |
468 | elevator_exit(q->elevator); | |
469 | ||
470 | blk_put_queue(q); | |
471 | } | |
1da177e4 LT |
472 | EXPORT_SYMBOL(blk_cleanup_queue); |
473 | ||
165125e1 | 474 | static int blk_init_free_list(struct request_queue *q) |
1da177e4 LT |
475 | { |
476 | struct request_list *rl = &q->rq; | |
477 | ||
1faa16d2 JA |
478 | rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; |
479 | rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; | |
cb98fc8b | 480 | rl->elvpriv = 0; |
1faa16d2 JA |
481 | init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); |
482 | init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); | |
1da177e4 | 483 | |
1946089a CL |
484 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
485 | mempool_free_slab, request_cachep, q->node); | |
1da177e4 LT |
486 | |
487 | if (!rl->rq_pool) | |
488 | return -ENOMEM; | |
489 | ||
490 | return 0; | |
491 | } | |
492 | ||
165125e1 | 493 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
1da177e4 | 494 | { |
1946089a CL |
495 | return blk_alloc_queue_node(gfp_mask, -1); |
496 | } | |
497 | EXPORT_SYMBOL(blk_alloc_queue); | |
1da177e4 | 498 | |
165125e1 | 499 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1946089a | 500 | { |
165125e1 | 501 | struct request_queue *q; |
e0bf68dd | 502 | int err; |
1946089a | 503 | |
8324aa91 | 504 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
94f6030c | 505 | gfp_mask | __GFP_ZERO, node_id); |
1da177e4 LT |
506 | if (!q) |
507 | return NULL; | |
508 | ||
e0bf68dd PZ |
509 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
510 | q->backing_dev_info.unplug_io_data = q; | |
511 | err = bdi_init(&q->backing_dev_info); | |
512 | if (err) { | |
8324aa91 | 513 | kmem_cache_free(blk_requestq_cachep, q); |
e0bf68dd PZ |
514 | return NULL; |
515 | } | |
516 | ||
1da177e4 | 517 | init_timer(&q->unplug_timer); |
242f9dcb JA |
518 | setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); |
519 | INIT_LIST_HEAD(&q->timeout_list); | |
713ada9b | 520 | INIT_WORK(&q->unplug_work, blk_unplug_work); |
483f4afc | 521 | |
8324aa91 | 522 | kobject_init(&q->kobj, &blk_queue_ktype); |
1da177e4 | 523 | |
483f4afc | 524 | mutex_init(&q->sysfs_lock); |
e7e72bf6 | 525 | spin_lock_init(&q->__queue_lock); |
483f4afc | 526 | |
1da177e4 LT |
527 | return q; |
528 | } | |
1946089a | 529 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1da177e4 LT |
530 | |
531 | /** | |
532 | * blk_init_queue - prepare a request queue for use with a block device | |
533 | * @rfn: The function to be called to process requests that have been | |
534 | * placed on the queue. | |
535 | * @lock: Request queue spin lock | |
536 | * | |
537 | * Description: | |
538 | * If a block device wishes to use the standard request handling procedures, | |
539 | * which sorts requests and coalesces adjacent requests, then it must | |
540 | * call blk_init_queue(). The function @rfn will be called when there | |
541 | * are requests on the queue that need to be processed. If the device | |
542 | * supports plugging, then @rfn may not be called immediately when requests | |
543 | * are available on the queue, but may be called at some time later instead. | |
544 | * Plugged queues are generally unplugged when a buffer belonging to one | |
545 | * of the requests on the queue is needed, or due to memory pressure. | |
546 | * | |
547 | * @rfn is not required, or even expected, to remove all requests off the | |
548 | * queue, but only as many as it can handle at a time. If it does leave | |
549 | * requests on the queue, it is responsible for arranging that the requests | |
550 | * get dealt with eventually. | |
551 | * | |
552 | * The queue spin lock must be held while manipulating the requests on the | |
a038e253 PBG |
553 | * request queue; this lock will be taken also from interrupt context, so irq |
554 | * disabling is needed for it. | |
1da177e4 | 555 | * |
710027a4 | 556 | * Function returns a pointer to the initialized request queue, or %NULL if |
1da177e4 LT |
557 | * it didn't succeed. |
558 | * | |
559 | * Note: | |
560 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | |
561 | * when the block device is deactivated (such as at module unload). | |
562 | **/ | |
1946089a | 563 | |
165125e1 | 564 | struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1da177e4 | 565 | { |
1946089a CL |
566 | return blk_init_queue_node(rfn, lock, -1); |
567 | } | |
568 | EXPORT_SYMBOL(blk_init_queue); | |
569 | ||
165125e1 | 570 | struct request_queue * |
1946089a CL |
571 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
572 | { | |
165125e1 | 573 | struct request_queue *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1da177e4 LT |
574 | |
575 | if (!q) | |
576 | return NULL; | |
577 | ||
1946089a | 578 | q->node = node_id; |
8669aafd | 579 | if (blk_init_free_list(q)) { |
8324aa91 | 580 | kmem_cache_free(blk_requestq_cachep, q); |
8669aafd AV |
581 | return NULL; |
582 | } | |
1da177e4 | 583 | |
152587de | 584 | /* |
585 | * if caller didn't supply a lock, they get per-queue locking with | |
586 | * our embedded lock | |
587 | */ | |
e7e72bf6 | 588 | if (!lock) |
152587de | 589 | lock = &q->__queue_lock; |
152587de | 590 | |
1da177e4 | 591 | q->request_fn = rfn; |
1da177e4 LT |
592 | q->prep_rq_fn = NULL; |
593 | q->unplug_fn = generic_unplug_device; | |
bc58ba94 | 594 | q->queue_flags = QUEUE_FLAG_DEFAULT; |
1da177e4 LT |
595 | q->queue_lock = lock; |
596 | ||
f3b144aa JA |
597 | /* |
598 | * This also sets hw/phys segments, boundary and size | |
599 | */ | |
1da177e4 | 600 | blk_queue_make_request(q, __make_request); |
1da177e4 | 601 | |
44ec9542 AS |
602 | q->sg_reserved_size = INT_MAX; |
603 | ||
abf54393 FT |
604 | blk_set_cmd_filter_defaults(&q->cmd_filter); |
605 | ||
1da177e4 LT |
606 | /* |
607 | * all done | |
608 | */ | |
609 | if (!elevator_init(q, NULL)) { | |
610 | blk_queue_congestion_threshold(q); | |
611 | return q; | |
612 | } | |
613 | ||
8669aafd | 614 | blk_put_queue(q); |
1da177e4 LT |
615 | return NULL; |
616 | } | |
1946089a | 617 | EXPORT_SYMBOL(blk_init_queue_node); |
1da177e4 | 618 | |
165125e1 | 619 | int blk_get_queue(struct request_queue *q) |
1da177e4 | 620 | { |
fde6ad22 | 621 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
483f4afc | 622 | kobject_get(&q->kobj); |
1da177e4 LT |
623 | return 0; |
624 | } | |
625 | ||
626 | return 1; | |
627 | } | |
1da177e4 | 628 | |
165125e1 | 629 | static inline void blk_free_request(struct request_queue *q, struct request *rq) |
1da177e4 | 630 | { |
4aff5e23 | 631 | if (rq->cmd_flags & REQ_ELVPRIV) |
cb98fc8b | 632 | elv_put_request(q, rq); |
1da177e4 LT |
633 | mempool_free(rq, q->rq.rq_pool); |
634 | } | |
635 | ||
1ea25ecb | 636 | static struct request * |
42dad764 | 637 | blk_alloc_request(struct request_queue *q, int flags, int priv, gfp_t gfp_mask) |
1da177e4 LT |
638 | { |
639 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | |
640 | ||
641 | if (!rq) | |
642 | return NULL; | |
643 | ||
2a4aa30c | 644 | blk_rq_init(q, rq); |
1afb20f3 | 645 | |
42dad764 | 646 | rq->cmd_flags = flags | REQ_ALLOCED; |
1da177e4 | 647 | |
cb98fc8b | 648 | if (priv) { |
cb78b285 | 649 | if (unlikely(elv_set_request(q, rq, gfp_mask))) { |
cb98fc8b TH |
650 | mempool_free(rq, q->rq.rq_pool); |
651 | return NULL; | |
652 | } | |
4aff5e23 | 653 | rq->cmd_flags |= REQ_ELVPRIV; |
cb98fc8b | 654 | } |
1da177e4 | 655 | |
cb98fc8b | 656 | return rq; |
1da177e4 LT |
657 | } |
658 | ||
659 | /* | |
660 | * ioc_batching returns true if the ioc is a valid batching request and | |
661 | * should be given priority access to a request. | |
662 | */ | |
165125e1 | 663 | static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
664 | { |
665 | if (!ioc) | |
666 | return 0; | |
667 | ||
668 | /* | |
669 | * Make sure the process is able to allocate at least 1 request | |
670 | * even if the batch times out, otherwise we could theoretically | |
671 | * lose wakeups. | |
672 | */ | |
673 | return ioc->nr_batch_requests == q->nr_batching || | |
674 | (ioc->nr_batch_requests > 0 | |
675 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | |
676 | } | |
677 | ||
678 | /* | |
679 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | |
680 | * will cause the process to be a "batcher" on all queues in the system. This | |
681 | * is the behaviour we want though - once it gets a wakeup it should be given | |
682 | * a nice run. | |
683 | */ | |
165125e1 | 684 | static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) |
1da177e4 LT |
685 | { |
686 | if (!ioc || ioc_batching(q, ioc)) | |
687 | return; | |
688 | ||
689 | ioc->nr_batch_requests = q->nr_batching; | |
690 | ioc->last_waited = jiffies; | |
691 | } | |
692 | ||
1faa16d2 | 693 | static void __freed_request(struct request_queue *q, int sync) |
1da177e4 LT |
694 | { |
695 | struct request_list *rl = &q->rq; | |
696 | ||
1faa16d2 JA |
697 | if (rl->count[sync] < queue_congestion_off_threshold(q)) |
698 | blk_clear_queue_congested(q, sync); | |
1da177e4 | 699 | |
1faa16d2 JA |
700 | if (rl->count[sync] + 1 <= q->nr_requests) { |
701 | if (waitqueue_active(&rl->wait[sync])) | |
702 | wake_up(&rl->wait[sync]); | |
1da177e4 | 703 | |
1faa16d2 | 704 | blk_clear_queue_full(q, sync); |
1da177e4 LT |
705 | } |
706 | } | |
707 | ||
708 | /* | |
709 | * A request has just been released. Account for it, update the full and | |
710 | * congestion status, wake up any waiters. Called under q->queue_lock. | |
711 | */ | |
1faa16d2 | 712 | static void freed_request(struct request_queue *q, int sync, int priv) |
1da177e4 LT |
713 | { |
714 | struct request_list *rl = &q->rq; | |
715 | ||
1faa16d2 | 716 | rl->count[sync]--; |
cb98fc8b TH |
717 | if (priv) |
718 | rl->elvpriv--; | |
1da177e4 | 719 | |
1faa16d2 | 720 | __freed_request(q, sync); |
1da177e4 | 721 | |
1faa16d2 JA |
722 | if (unlikely(rl->starved[sync ^ 1])) |
723 | __freed_request(q, sync ^ 1); | |
1da177e4 LT |
724 | } |
725 | ||
1da177e4 | 726 | /* |
d6344532 NP |
727 | * Get a free request, queue_lock must be held. |
728 | * Returns NULL on failure, with queue_lock held. | |
729 | * Returns !NULL on success, with queue_lock *not held*. | |
1da177e4 | 730 | */ |
165125e1 | 731 | static struct request *get_request(struct request_queue *q, int rw_flags, |
7749a8d4 | 732 | struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
733 | { |
734 | struct request *rq = NULL; | |
735 | struct request_list *rl = &q->rq; | |
88ee5ef1 | 736 | struct io_context *ioc = NULL; |
1faa16d2 | 737 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
88ee5ef1 JA |
738 | int may_queue, priv; |
739 | ||
7749a8d4 | 740 | may_queue = elv_may_queue(q, rw_flags); |
88ee5ef1 JA |
741 | if (may_queue == ELV_MQUEUE_NO) |
742 | goto rq_starved; | |
743 | ||
1faa16d2 JA |
744 | if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { |
745 | if (rl->count[is_sync]+1 >= q->nr_requests) { | |
b5deef90 | 746 | ioc = current_io_context(GFP_ATOMIC, q->node); |
88ee5ef1 JA |
747 | /* |
748 | * The queue will fill after this allocation, so set | |
749 | * it as full, and mark this process as "batching". | |
750 | * This process will be allowed to complete a batch of | |
751 | * requests, others will be blocked. | |
752 | */ | |
1faa16d2 | 753 | if (!blk_queue_full(q, is_sync)) { |
88ee5ef1 | 754 | ioc_set_batching(q, ioc); |
1faa16d2 | 755 | blk_set_queue_full(q, is_sync); |
88ee5ef1 JA |
756 | } else { |
757 | if (may_queue != ELV_MQUEUE_MUST | |
758 | && !ioc_batching(q, ioc)) { | |
759 | /* | |
760 | * The queue is full and the allocating | |
761 | * process is not a "batcher", and not | |
762 | * exempted by the IO scheduler | |
763 | */ | |
764 | goto out; | |
765 | } | |
766 | } | |
1da177e4 | 767 | } |
1faa16d2 | 768 | blk_set_queue_congested(q, is_sync); |
1da177e4 LT |
769 | } |
770 | ||
082cf69e JA |
771 | /* |
772 | * Only allow batching queuers to allocate up to 50% over the defined | |
773 | * limit of requests, otherwise we could have thousands of requests | |
774 | * allocated with any setting of ->nr_requests | |
775 | */ | |
1faa16d2 | 776 | if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) |
082cf69e | 777 | goto out; |
fd782a4a | 778 | |
1faa16d2 JA |
779 | rl->count[is_sync]++; |
780 | rl->starved[is_sync] = 0; | |
cb98fc8b | 781 | |
64521d1a | 782 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
783 | if (priv) |
784 | rl->elvpriv++; | |
785 | ||
42dad764 JM |
786 | if (blk_queue_io_stat(q)) |
787 | rw_flags |= REQ_IO_STAT; | |
1da177e4 LT |
788 | spin_unlock_irq(q->queue_lock); |
789 | ||
7749a8d4 | 790 | rq = blk_alloc_request(q, rw_flags, priv, gfp_mask); |
88ee5ef1 | 791 | if (unlikely(!rq)) { |
1da177e4 LT |
792 | /* |
793 | * Allocation failed presumably due to memory. Undo anything | |
794 | * we might have messed up. | |
795 | * | |
796 | * Allocating task should really be put onto the front of the | |
797 | * wait queue, but this is pretty rare. | |
798 | */ | |
799 | spin_lock_irq(q->queue_lock); | |
1faa16d2 | 800 | freed_request(q, is_sync, priv); |
1da177e4 LT |
801 | |
802 | /* | |
803 | * in the very unlikely event that allocation failed and no | |
804 | * requests for this direction was pending, mark us starved | |
805 | * so that freeing of a request in the other direction will | |
806 | * notice us. another possible fix would be to split the | |
807 | * rq mempool into READ and WRITE | |
808 | */ | |
809 | rq_starved: | |
1faa16d2 JA |
810 | if (unlikely(rl->count[is_sync] == 0)) |
811 | rl->starved[is_sync] = 1; | |
1da177e4 | 812 | |
1da177e4 LT |
813 | goto out; |
814 | } | |
815 | ||
88ee5ef1 JA |
816 | /* |
817 | * ioc may be NULL here, and ioc_batching will be false. That's | |
818 | * OK, if the queue is under the request limit then requests need | |
819 | * not count toward the nr_batch_requests limit. There will always | |
820 | * be some limit enforced by BLK_BATCH_TIME. | |
821 | */ | |
1da177e4 LT |
822 | if (ioc_batching(q, ioc)) |
823 | ioc->nr_batch_requests--; | |
6728cb0e | 824 | |
1faa16d2 | 825 | trace_block_getrq(q, bio, rw_flags & 1); |
1da177e4 | 826 | out: |
1da177e4 LT |
827 | return rq; |
828 | } | |
829 | ||
830 | /* | |
831 | * No available requests for this queue, unplug the device and wait for some | |
832 | * requests to become available. | |
d6344532 NP |
833 | * |
834 | * Called with q->queue_lock held, and returns with it unlocked. | |
1da177e4 | 835 | */ |
165125e1 | 836 | static struct request *get_request_wait(struct request_queue *q, int rw_flags, |
22e2c507 | 837 | struct bio *bio) |
1da177e4 | 838 | { |
1faa16d2 | 839 | const bool is_sync = rw_is_sync(rw_flags) != 0; |
1da177e4 LT |
840 | struct request *rq; |
841 | ||
7749a8d4 | 842 | rq = get_request(q, rw_flags, bio, GFP_NOIO); |
450991bc NP |
843 | while (!rq) { |
844 | DEFINE_WAIT(wait); | |
05caf8db | 845 | struct io_context *ioc; |
1da177e4 LT |
846 | struct request_list *rl = &q->rq; |
847 | ||
1faa16d2 | 848 | prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, |
1da177e4 LT |
849 | TASK_UNINTERRUPTIBLE); |
850 | ||
1faa16d2 | 851 | trace_block_sleeprq(q, bio, rw_flags & 1); |
1da177e4 | 852 | |
05caf8db ZY |
853 | __generic_unplug_device(q); |
854 | spin_unlock_irq(q->queue_lock); | |
855 | io_schedule(); | |
1da177e4 | 856 | |
05caf8db ZY |
857 | /* |
858 | * After sleeping, we become a "batching" process and | |
859 | * will be able to allocate at least one request, and | |
860 | * up to a big batch of them for a small period time. | |
861 | * See ioc_batching, ioc_set_batching | |
862 | */ | |
863 | ioc = current_io_context(GFP_NOIO, q->node); | |
864 | ioc_set_batching(q, ioc); | |
d6344532 | 865 | |
05caf8db | 866 | spin_lock_irq(q->queue_lock); |
1faa16d2 | 867 | finish_wait(&rl->wait[is_sync], &wait); |
05caf8db ZY |
868 | |
869 | rq = get_request(q, rw_flags, bio, GFP_NOIO); | |
870 | }; | |
1da177e4 LT |
871 | |
872 | return rq; | |
873 | } | |
874 | ||
165125e1 | 875 | struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) |
1da177e4 LT |
876 | { |
877 | struct request *rq; | |
878 | ||
879 | BUG_ON(rw != READ && rw != WRITE); | |
880 | ||
d6344532 NP |
881 | spin_lock_irq(q->queue_lock); |
882 | if (gfp_mask & __GFP_WAIT) { | |
22e2c507 | 883 | rq = get_request_wait(q, rw, NULL); |
d6344532 | 884 | } else { |
22e2c507 | 885 | rq = get_request(q, rw, NULL, gfp_mask); |
d6344532 NP |
886 | if (!rq) |
887 | spin_unlock_irq(q->queue_lock); | |
888 | } | |
889 | /* q->queue_lock is unlocked at this point */ | |
1da177e4 LT |
890 | |
891 | return rq; | |
892 | } | |
1da177e4 LT |
893 | EXPORT_SYMBOL(blk_get_request); |
894 | ||
895 | /** | |
896 | * blk_requeue_request - put a request back on queue | |
897 | * @q: request queue where request should be inserted | |
898 | * @rq: request to be inserted | |
899 | * | |
900 | * Description: | |
901 | * Drivers often keep queueing requests until the hardware cannot accept | |
902 | * more, when that condition happens we need to put the request back | |
903 | * on the queue. Must be called with queue lock held. | |
904 | */ | |
165125e1 | 905 | void blk_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 906 | { |
242f9dcb JA |
907 | blk_delete_timer(rq); |
908 | blk_clear_rq_complete(rq); | |
5f3ea37c | 909 | trace_block_rq_requeue(q, rq); |
2056a782 | 910 | |
1da177e4 LT |
911 | if (blk_rq_tagged(rq)) |
912 | blk_queue_end_tag(q, rq); | |
913 | ||
914 | elv_requeue_request(q, rq); | |
915 | } | |
1da177e4 LT |
916 | EXPORT_SYMBOL(blk_requeue_request); |
917 | ||
918 | /** | |
710027a4 | 919 | * blk_insert_request - insert a special request into a request queue |
1da177e4 LT |
920 | * @q: request queue where request should be inserted |
921 | * @rq: request to be inserted | |
922 | * @at_head: insert request at head or tail of queue | |
923 | * @data: private data | |
1da177e4 LT |
924 | * |
925 | * Description: | |
926 | * Many block devices need to execute commands asynchronously, so they don't | |
927 | * block the whole kernel from preemption during request execution. This is | |
928 | * accomplished normally by inserting aritficial requests tagged as | |
710027a4 RD |
929 | * REQ_TYPE_SPECIAL in to the corresponding request queue, and letting them |
930 | * be scheduled for actual execution by the request queue. | |
1da177e4 LT |
931 | * |
932 | * We have the option of inserting the head or the tail of the queue. | |
933 | * Typically we use the tail for new ioctls and so forth. We use the head | |
934 | * of the queue for things like a QUEUE_FULL message from a device, or a | |
935 | * host that is unable to accept a particular command. | |
936 | */ | |
165125e1 | 937 | void blk_insert_request(struct request_queue *q, struct request *rq, |
867d1191 | 938 | int at_head, void *data) |
1da177e4 | 939 | { |
867d1191 | 940 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
1da177e4 LT |
941 | unsigned long flags; |
942 | ||
943 | /* | |
944 | * tell I/O scheduler that this isn't a regular read/write (ie it | |
945 | * must not attempt merges on this) and that it acts as a soft | |
946 | * barrier | |
947 | */ | |
4aff5e23 | 948 | rq->cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 LT |
949 | |
950 | rq->special = data; | |
951 | ||
952 | spin_lock_irqsave(q->queue_lock, flags); | |
953 | ||
954 | /* | |
955 | * If command is tagged, release the tag | |
956 | */ | |
867d1191 TH |
957 | if (blk_rq_tagged(rq)) |
958 | blk_queue_end_tag(q, rq); | |
1da177e4 | 959 | |
b238b3d4 | 960 | drive_stat_acct(rq, 1); |
867d1191 | 961 | __elv_add_request(q, rq, where, 0); |
a7f55792 | 962 | __blk_run_queue(q); |
1da177e4 LT |
963 | spin_unlock_irqrestore(q->queue_lock, flags); |
964 | } | |
1da177e4 LT |
965 | EXPORT_SYMBOL(blk_insert_request); |
966 | ||
1da177e4 LT |
967 | /* |
968 | * add-request adds a request to the linked list. | |
969 | * queue lock is held and interrupts disabled, as we muck with the | |
970 | * request queue list. | |
971 | */ | |
6728cb0e | 972 | static inline void add_request(struct request_queue *q, struct request *req) |
1da177e4 | 973 | { |
b238b3d4 | 974 | drive_stat_acct(req, 1); |
1da177e4 | 975 | |
1da177e4 LT |
976 | /* |
977 | * elevator indicated where it wants this request to be | |
978 | * inserted at elevator_merge time | |
979 | */ | |
980 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | |
981 | } | |
6728cb0e | 982 | |
074a7aca TH |
983 | static void part_round_stats_single(int cpu, struct hd_struct *part, |
984 | unsigned long now) | |
985 | { | |
986 | if (now == part->stamp) | |
987 | return; | |
988 | ||
989 | if (part->in_flight) { | |
990 | __part_stat_add(cpu, part, time_in_queue, | |
991 | part->in_flight * (now - part->stamp)); | |
992 | __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); | |
993 | } | |
994 | part->stamp = now; | |
995 | } | |
996 | ||
997 | /** | |
496aa8a9 RD |
998 | * part_round_stats() - Round off the performance stats on a struct disk_stats. |
999 | * @cpu: cpu number for stats access | |
1000 | * @part: target partition | |
1da177e4 LT |
1001 | * |
1002 | * The average IO queue length and utilisation statistics are maintained | |
1003 | * by observing the current state of the queue length and the amount of | |
1004 | * time it has been in this state for. | |
1005 | * | |
1006 | * Normally, that accounting is done on IO completion, but that can result | |
1007 | * in more than a second's worth of IO being accounted for within any one | |
1008 | * second, leading to >100% utilisation. To deal with that, we call this | |
1009 | * function to do a round-off before returning the results when reading | |
1010 | * /proc/diskstats. This accounts immediately for all queue usage up to | |
1011 | * the current jiffies and restarts the counters again. | |
1012 | */ | |
c9959059 | 1013 | void part_round_stats(int cpu, struct hd_struct *part) |
6f2576af JM |
1014 | { |
1015 | unsigned long now = jiffies; | |
1016 | ||
074a7aca TH |
1017 | if (part->partno) |
1018 | part_round_stats_single(cpu, &part_to_disk(part)->part0, now); | |
1019 | part_round_stats_single(cpu, part, now); | |
6f2576af | 1020 | } |
074a7aca | 1021 | EXPORT_SYMBOL_GPL(part_round_stats); |
6f2576af | 1022 | |
1da177e4 LT |
1023 | /* |
1024 | * queue lock must be held | |
1025 | */ | |
165125e1 | 1026 | void __blk_put_request(struct request_queue *q, struct request *req) |
1da177e4 | 1027 | { |
1da177e4 LT |
1028 | if (unlikely(!q)) |
1029 | return; | |
1030 | if (unlikely(--req->ref_count)) | |
1031 | return; | |
1032 | ||
8922e16c TH |
1033 | elv_completed_request(q, req); |
1034 | ||
1cd96c24 BH |
1035 | /* this is a bio leak */ |
1036 | WARN_ON(req->bio != NULL); | |
1037 | ||
1da177e4 LT |
1038 | /* |
1039 | * Request may not have originated from ll_rw_blk. if not, | |
1040 | * it didn't come out of our reserved rq pools | |
1041 | */ | |
49171e5c | 1042 | if (req->cmd_flags & REQ_ALLOCED) { |
1faa16d2 | 1043 | int is_sync = rq_is_sync(req) != 0; |
4aff5e23 | 1044 | int priv = req->cmd_flags & REQ_ELVPRIV; |
1da177e4 | 1045 | |
1da177e4 | 1046 | BUG_ON(!list_empty(&req->queuelist)); |
9817064b | 1047 | BUG_ON(!hlist_unhashed(&req->hash)); |
1da177e4 LT |
1048 | |
1049 | blk_free_request(q, req); | |
1faa16d2 | 1050 | freed_request(q, is_sync, priv); |
1da177e4 LT |
1051 | } |
1052 | } | |
6e39b69e MC |
1053 | EXPORT_SYMBOL_GPL(__blk_put_request); |
1054 | ||
1da177e4 LT |
1055 | void blk_put_request(struct request *req) |
1056 | { | |
8922e16c | 1057 | unsigned long flags; |
165125e1 | 1058 | struct request_queue *q = req->q; |
8922e16c | 1059 | |
52a93ba8 FT |
1060 | spin_lock_irqsave(q->queue_lock, flags); |
1061 | __blk_put_request(q, req); | |
1062 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1da177e4 | 1063 | } |
1da177e4 LT |
1064 | EXPORT_SYMBOL(blk_put_request); |
1065 | ||
86db1e29 | 1066 | void init_request_from_bio(struct request *req, struct bio *bio) |
52d9e675 | 1067 | { |
c7c22e4d | 1068 | req->cpu = bio->bi_comp_cpu; |
4aff5e23 | 1069 | req->cmd_type = REQ_TYPE_FS; |
52d9e675 TH |
1070 | |
1071 | /* | |
1072 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | |
1073 | */ | |
6000a368 MC |
1074 | if (bio_rw_ahead(bio)) |
1075 | req->cmd_flags |= (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT | | |
1076 | REQ_FAILFAST_DRIVER); | |
1077 | if (bio_failfast_dev(bio)) | |
1078 | req->cmd_flags |= REQ_FAILFAST_DEV; | |
1079 | if (bio_failfast_transport(bio)) | |
1080 | req->cmd_flags |= REQ_FAILFAST_TRANSPORT; | |
1081 | if (bio_failfast_driver(bio)) | |
1082 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
52d9e675 | 1083 | |
fb2dce86 | 1084 | if (unlikely(bio_discard(bio))) { |
e17fc0a1 DW |
1085 | req->cmd_flags |= REQ_DISCARD; |
1086 | if (bio_barrier(bio)) | |
1087 | req->cmd_flags |= REQ_SOFTBARRIER; | |
fb2dce86 | 1088 | req->q->prepare_discard_fn(req->q, req); |
e17fc0a1 | 1089 | } else if (unlikely(bio_barrier(bio))) |
e4025f6c | 1090 | req->cmd_flags |= REQ_HARDBARRIER; |
52d9e675 | 1091 | |
b31dc66a | 1092 | if (bio_sync(bio)) |
4aff5e23 | 1093 | req->cmd_flags |= REQ_RW_SYNC; |
5404bc7a JA |
1094 | if (bio_rw_meta(bio)) |
1095 | req->cmd_flags |= REQ_RW_META; | |
aeb6fafb JA |
1096 | if (bio_noidle(bio)) |
1097 | req->cmd_flags |= REQ_NOIDLE; | |
b31dc66a | 1098 | |
52d9e675 TH |
1099 | req->errors = 0; |
1100 | req->hard_sector = req->sector = bio->bi_sector; | |
52d9e675 | 1101 | req->ioprio = bio_prio(bio); |
52d9e675 | 1102 | req->start_time = jiffies; |
bc1c56fd | 1103 | blk_rq_bio_prep(req->q, req, bio); |
52d9e675 TH |
1104 | } |
1105 | ||
644b2d99 JA |
1106 | /* |
1107 | * Only disabling plugging for non-rotational devices if it does tagging | |
1108 | * as well, otherwise we do need the proper merging | |
1109 | */ | |
1110 | static inline bool queue_should_plug(struct request_queue *q) | |
1111 | { | |
1112 | return !(blk_queue_nonrot(q) && blk_queue_tagged(q)); | |
1113 | } | |
1114 | ||
165125e1 | 1115 | static int __make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1116 | { |
450991bc | 1117 | struct request *req; |
a7384677 | 1118 | int el_ret, nr_sectors; |
51da90fc JA |
1119 | const unsigned short prio = bio_prio(bio); |
1120 | const int sync = bio_sync(bio); | |
213d9417 | 1121 | const int unplug = bio_unplug(bio); |
7749a8d4 | 1122 | int rw_flags; |
1da177e4 | 1123 | |
1da177e4 | 1124 | nr_sectors = bio_sectors(bio); |
1da177e4 LT |
1125 | |
1126 | /* | |
1127 | * low level driver can indicate that it wants pages above a | |
1128 | * certain limit bounced to low memory (ie for highmem, or even | |
1129 | * ISA dma in theory) | |
1130 | */ | |
1131 | blk_queue_bounce(q, &bio); | |
1132 | ||
1da177e4 LT |
1133 | spin_lock_irq(q->queue_lock); |
1134 | ||
a7384677 | 1135 | if (unlikely(bio_barrier(bio)) || elv_queue_empty(q)) |
1da177e4 LT |
1136 | goto get_rq; |
1137 | ||
1138 | el_ret = elv_merge(q, &req, bio); | |
1139 | switch (el_ret) { | |
6728cb0e JA |
1140 | case ELEVATOR_BACK_MERGE: |
1141 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1142 | |
6728cb0e JA |
1143 | if (!ll_back_merge_fn(q, req, bio)) |
1144 | break; | |
1da177e4 | 1145 | |
5f3ea37c | 1146 | trace_block_bio_backmerge(q, bio); |
2056a782 | 1147 | |
6728cb0e JA |
1148 | req->biotail->bi_next = bio; |
1149 | req->biotail = bio; | |
1150 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1151 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1152 | if (!blk_rq_cpu_valid(req)) |
1153 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1154 | drive_stat_acct(req, 0); |
1155 | if (!attempt_back_merge(q, req)) | |
1156 | elv_merged_request(q, req, el_ret); | |
1157 | goto out; | |
1da177e4 | 1158 | |
6728cb0e JA |
1159 | case ELEVATOR_FRONT_MERGE: |
1160 | BUG_ON(!rq_mergeable(req)); | |
1da177e4 | 1161 | |
6728cb0e JA |
1162 | if (!ll_front_merge_fn(q, req, bio)) |
1163 | break; | |
1da177e4 | 1164 | |
5f3ea37c | 1165 | trace_block_bio_frontmerge(q, bio); |
2056a782 | 1166 | |
6728cb0e JA |
1167 | bio->bi_next = req->bio; |
1168 | req->bio = bio; | |
1da177e4 | 1169 | |
6728cb0e JA |
1170 | /* |
1171 | * may not be valid. if the low level driver said | |
1172 | * it didn't need a bounce buffer then it better | |
1173 | * not touch req->buffer either... | |
1174 | */ | |
1175 | req->buffer = bio_data(bio); | |
1176 | req->current_nr_sectors = bio_cur_sectors(bio); | |
1177 | req->hard_cur_sectors = req->current_nr_sectors; | |
1178 | req->sector = req->hard_sector = bio->bi_sector; | |
1179 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | |
1180 | req->ioprio = ioprio_best(req->ioprio, prio); | |
ab780f1e JA |
1181 | if (!blk_rq_cpu_valid(req)) |
1182 | req->cpu = bio->bi_comp_cpu; | |
6728cb0e JA |
1183 | drive_stat_acct(req, 0); |
1184 | if (!attempt_front_merge(q, req)) | |
1185 | elv_merged_request(q, req, el_ret); | |
1186 | goto out; | |
1187 | ||
1188 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | |
1189 | default: | |
1190 | ; | |
1da177e4 LT |
1191 | } |
1192 | ||
450991bc | 1193 | get_rq: |
7749a8d4 JA |
1194 | /* |
1195 | * This sync check and mask will be re-done in init_request_from_bio(), | |
1196 | * but we need to set it earlier to expose the sync flag to the | |
1197 | * rq allocator and io schedulers. | |
1198 | */ | |
1199 | rw_flags = bio_data_dir(bio); | |
1200 | if (sync) | |
1201 | rw_flags |= REQ_RW_SYNC; | |
1202 | ||
1da177e4 | 1203 | /* |
450991bc | 1204 | * Grab a free request. This is might sleep but can not fail. |
d6344532 | 1205 | * Returns with the queue unlocked. |
450991bc | 1206 | */ |
7749a8d4 | 1207 | req = get_request_wait(q, rw_flags, bio); |
d6344532 | 1208 | |
450991bc NP |
1209 | /* |
1210 | * After dropping the lock and possibly sleeping here, our request | |
1211 | * may now be mergeable after it had proven unmergeable (above). | |
1212 | * We don't worry about that case for efficiency. It won't happen | |
1213 | * often, and the elevators are able to handle it. | |
1da177e4 | 1214 | */ |
52d9e675 | 1215 | init_request_from_bio(req, bio); |
1da177e4 | 1216 | |
450991bc | 1217 | spin_lock_irq(q->queue_lock); |
c7c22e4d JA |
1218 | if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags) || |
1219 | bio_flagged(bio, BIO_CPU_AFFINE)) | |
1220 | req->cpu = blk_cpu_to_group(smp_processor_id()); | |
644b2d99 | 1221 | if (queue_should_plug(q) && elv_queue_empty(q)) |
450991bc | 1222 | blk_plug_device(q); |
1da177e4 LT |
1223 | add_request(q, req); |
1224 | out: | |
644b2d99 | 1225 | if (unplug || !queue_should_plug(q)) |
1da177e4 | 1226 | __generic_unplug_device(q); |
1da177e4 LT |
1227 | spin_unlock_irq(q->queue_lock); |
1228 | return 0; | |
1da177e4 LT |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * If bio->bi_dev is a partition, remap the location | |
1233 | */ | |
1234 | static inline void blk_partition_remap(struct bio *bio) | |
1235 | { | |
1236 | struct block_device *bdev = bio->bi_bdev; | |
1237 | ||
bf2de6f5 | 1238 | if (bio_sectors(bio) && bdev != bdev->bd_contains) { |
1da177e4 LT |
1239 | struct hd_struct *p = bdev->bd_part; |
1240 | ||
1da177e4 LT |
1241 | bio->bi_sector += p->start_sect; |
1242 | bio->bi_bdev = bdev->bd_contains; | |
c7149d6b | 1243 | |
5f3ea37c | 1244 | trace_block_remap(bdev_get_queue(bio->bi_bdev), bio, |
c7149d6b AB |
1245 | bdev->bd_dev, bio->bi_sector, |
1246 | bio->bi_sector - p->start_sect); | |
1da177e4 LT |
1247 | } |
1248 | } | |
1249 | ||
1da177e4 LT |
1250 | static void handle_bad_sector(struct bio *bio) |
1251 | { | |
1252 | char b[BDEVNAME_SIZE]; | |
1253 | ||
1254 | printk(KERN_INFO "attempt to access beyond end of device\n"); | |
1255 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | |
1256 | bdevname(bio->bi_bdev, b), | |
1257 | bio->bi_rw, | |
1258 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | |
1259 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | |
1260 | ||
1261 | set_bit(BIO_EOF, &bio->bi_flags); | |
1262 | } | |
1263 | ||
c17bb495 AM |
1264 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
1265 | ||
1266 | static DECLARE_FAULT_ATTR(fail_make_request); | |
1267 | ||
1268 | static int __init setup_fail_make_request(char *str) | |
1269 | { | |
1270 | return setup_fault_attr(&fail_make_request, str); | |
1271 | } | |
1272 | __setup("fail_make_request=", setup_fail_make_request); | |
1273 | ||
1274 | static int should_fail_request(struct bio *bio) | |
1275 | { | |
eddb2e26 TH |
1276 | struct hd_struct *part = bio->bi_bdev->bd_part; |
1277 | ||
1278 | if (part_to_disk(part)->part0.make_it_fail || part->make_it_fail) | |
c17bb495 AM |
1279 | return should_fail(&fail_make_request, bio->bi_size); |
1280 | ||
1281 | return 0; | |
1282 | } | |
1283 | ||
1284 | static int __init fail_make_request_debugfs(void) | |
1285 | { | |
1286 | return init_fault_attr_dentries(&fail_make_request, | |
1287 | "fail_make_request"); | |
1288 | } | |
1289 | ||
1290 | late_initcall(fail_make_request_debugfs); | |
1291 | ||
1292 | #else /* CONFIG_FAIL_MAKE_REQUEST */ | |
1293 | ||
1294 | static inline int should_fail_request(struct bio *bio) | |
1295 | { | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ | |
1300 | ||
c07e2b41 JA |
1301 | /* |
1302 | * Check whether this bio extends beyond the end of the device. | |
1303 | */ | |
1304 | static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) | |
1305 | { | |
1306 | sector_t maxsector; | |
1307 | ||
1308 | if (!nr_sectors) | |
1309 | return 0; | |
1310 | ||
1311 | /* Test device or partition size, when known. */ | |
1312 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | |
1313 | if (maxsector) { | |
1314 | sector_t sector = bio->bi_sector; | |
1315 | ||
1316 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | |
1317 | /* | |
1318 | * This may well happen - the kernel calls bread() | |
1319 | * without checking the size of the device, e.g., when | |
1320 | * mounting a device. | |
1321 | */ | |
1322 | handle_bad_sector(bio); | |
1323 | return 1; | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | return 0; | |
1328 | } | |
1329 | ||
1da177e4 | 1330 | /** |
710027a4 | 1331 | * generic_make_request - hand a buffer to its device driver for I/O |
1da177e4 LT |
1332 | * @bio: The bio describing the location in memory and on the device. |
1333 | * | |
1334 | * generic_make_request() is used to make I/O requests of block | |
1335 | * devices. It is passed a &struct bio, which describes the I/O that needs | |
1336 | * to be done. | |
1337 | * | |
1338 | * generic_make_request() does not return any status. The | |
1339 | * success/failure status of the request, along with notification of | |
1340 | * completion, is delivered asynchronously through the bio->bi_end_io | |
1341 | * function described (one day) else where. | |
1342 | * | |
1343 | * The caller of generic_make_request must make sure that bi_io_vec | |
1344 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | |
1345 | * set to describe the device address, and the | |
1346 | * bi_end_io and optionally bi_private are set to describe how | |
1347 | * completion notification should be signaled. | |
1348 | * | |
1349 | * generic_make_request and the drivers it calls may use bi_next if this | |
1350 | * bio happens to be merged with someone else, and may change bi_dev and | |
1351 | * bi_sector for remaps as it sees fit. So the values of these fields | |
1352 | * should NOT be depended on after the call to generic_make_request. | |
1353 | */ | |
d89d8796 | 1354 | static inline void __generic_make_request(struct bio *bio) |
1da177e4 | 1355 | { |
165125e1 | 1356 | struct request_queue *q; |
5ddfe969 | 1357 | sector_t old_sector; |
1da177e4 | 1358 | int ret, nr_sectors = bio_sectors(bio); |
2056a782 | 1359 | dev_t old_dev; |
51fd77bd | 1360 | int err = -EIO; |
1da177e4 LT |
1361 | |
1362 | might_sleep(); | |
1da177e4 | 1363 | |
c07e2b41 JA |
1364 | if (bio_check_eod(bio, nr_sectors)) |
1365 | goto end_io; | |
1da177e4 LT |
1366 | |
1367 | /* | |
1368 | * Resolve the mapping until finished. (drivers are | |
1369 | * still free to implement/resolve their own stacking | |
1370 | * by explicitly returning 0) | |
1371 | * | |
1372 | * NOTE: we don't repeat the blk_size check for each new device. | |
1373 | * Stacking drivers are expected to know what they are doing. | |
1374 | */ | |
5ddfe969 | 1375 | old_sector = -1; |
2056a782 | 1376 | old_dev = 0; |
1da177e4 LT |
1377 | do { |
1378 | char b[BDEVNAME_SIZE]; | |
1379 | ||
1380 | q = bdev_get_queue(bio->bi_bdev); | |
a7384677 | 1381 | if (unlikely(!q)) { |
1da177e4 LT |
1382 | printk(KERN_ERR |
1383 | "generic_make_request: Trying to access " | |
1384 | "nonexistent block-device %s (%Lu)\n", | |
1385 | bdevname(bio->bi_bdev, b), | |
1386 | (long long) bio->bi_sector); | |
a7384677 | 1387 | goto end_io; |
1da177e4 LT |
1388 | } |
1389 | ||
4fa253f3 | 1390 | if (unlikely(nr_sectors > q->max_hw_sectors)) { |
6728cb0e | 1391 | printk(KERN_ERR "bio too big device %s (%u > %u)\n", |
1da177e4 LT |
1392 | bdevname(bio->bi_bdev, b), |
1393 | bio_sectors(bio), | |
1394 | q->max_hw_sectors); | |
1395 | goto end_io; | |
1396 | } | |
1397 | ||
fde6ad22 | 1398 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
1da177e4 LT |
1399 | goto end_io; |
1400 | ||
c17bb495 AM |
1401 | if (should_fail_request(bio)) |
1402 | goto end_io; | |
1403 | ||
1da177e4 LT |
1404 | /* |
1405 | * If this device has partitions, remap block n | |
1406 | * of partition p to block n+start(p) of the disk. | |
1407 | */ | |
1408 | blk_partition_remap(bio); | |
1409 | ||
7ba1ba12 MP |
1410 | if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) |
1411 | goto end_io; | |
1412 | ||
5ddfe969 | 1413 | if (old_sector != -1) |
5f3ea37c | 1414 | trace_block_remap(q, bio, old_dev, bio->bi_sector, |
5ddfe969 | 1415 | old_sector); |
2056a782 | 1416 | |
5f3ea37c | 1417 | trace_block_bio_queue(q, bio); |
2056a782 | 1418 | |
5ddfe969 | 1419 | old_sector = bio->bi_sector; |
2056a782 JA |
1420 | old_dev = bio->bi_bdev->bd_dev; |
1421 | ||
c07e2b41 JA |
1422 | if (bio_check_eod(bio, nr_sectors)) |
1423 | goto end_io; | |
a7384677 TH |
1424 | |
1425 | if (bio_discard(bio) && !q->prepare_discard_fn) { | |
51fd77bd JA |
1426 | err = -EOPNOTSUPP; |
1427 | goto end_io; | |
1428 | } | |
cec0707e JA |
1429 | if (bio_barrier(bio) && bio_has_data(bio) && |
1430 | (q->next_ordered == QUEUE_ORDERED_NONE)) { | |
1431 | err = -EOPNOTSUPP; | |
1432 | goto end_io; | |
1433 | } | |
5ddfe969 | 1434 | |
1da177e4 LT |
1435 | ret = q->make_request_fn(q, bio); |
1436 | } while (ret); | |
a7384677 TH |
1437 | |
1438 | return; | |
1439 | ||
1440 | end_io: | |
1441 | bio_endio(bio, err); | |
1da177e4 LT |
1442 | } |
1443 | ||
d89d8796 NB |
1444 | /* |
1445 | * We only want one ->make_request_fn to be active at a time, | |
1446 | * else stack usage with stacked devices could be a problem. | |
1447 | * So use current->bio_{list,tail} to keep a list of requests | |
1448 | * submited by a make_request_fn function. | |
1449 | * current->bio_tail is also used as a flag to say if | |
1450 | * generic_make_request is currently active in this task or not. | |
1451 | * If it is NULL, then no make_request is active. If it is non-NULL, | |
1452 | * then a make_request is active, and new requests should be added | |
1453 | * at the tail | |
1454 | */ | |
1455 | void generic_make_request(struct bio *bio) | |
1456 | { | |
1457 | if (current->bio_tail) { | |
1458 | /* make_request is active */ | |
1459 | *(current->bio_tail) = bio; | |
1460 | bio->bi_next = NULL; | |
1461 | current->bio_tail = &bio->bi_next; | |
1462 | return; | |
1463 | } | |
1464 | /* following loop may be a bit non-obvious, and so deserves some | |
1465 | * explanation. | |
1466 | * Before entering the loop, bio->bi_next is NULL (as all callers | |
1467 | * ensure that) so we have a list with a single bio. | |
1468 | * We pretend that we have just taken it off a longer list, so | |
1469 | * we assign bio_list to the next (which is NULL) and bio_tail | |
1470 | * to &bio_list, thus initialising the bio_list of new bios to be | |
1471 | * added. __generic_make_request may indeed add some more bios | |
1472 | * through a recursive call to generic_make_request. If it | |
1473 | * did, we find a non-NULL value in bio_list and re-enter the loop | |
1474 | * from the top. In this case we really did just take the bio | |
1475 | * of the top of the list (no pretending) and so fixup bio_list and | |
1476 | * bio_tail or bi_next, and call into __generic_make_request again. | |
1477 | * | |
1478 | * The loop was structured like this to make only one call to | |
1479 | * __generic_make_request (which is important as it is large and | |
1480 | * inlined) and to keep the structure simple. | |
1481 | */ | |
1482 | BUG_ON(bio->bi_next); | |
1483 | do { | |
1484 | current->bio_list = bio->bi_next; | |
1485 | if (bio->bi_next == NULL) | |
1486 | current->bio_tail = ¤t->bio_list; | |
1487 | else | |
1488 | bio->bi_next = NULL; | |
1489 | __generic_make_request(bio); | |
1490 | bio = current->bio_list; | |
1491 | } while (bio); | |
1492 | current->bio_tail = NULL; /* deactivate */ | |
1493 | } | |
1da177e4 LT |
1494 | EXPORT_SYMBOL(generic_make_request); |
1495 | ||
1496 | /** | |
710027a4 | 1497 | * submit_bio - submit a bio to the block device layer for I/O |
1da177e4 LT |
1498 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
1499 | * @bio: The &struct bio which describes the I/O | |
1500 | * | |
1501 | * submit_bio() is very similar in purpose to generic_make_request(), and | |
1502 | * uses that function to do most of the work. Both are fairly rough | |
710027a4 | 1503 | * interfaces; @bio must be presetup and ready for I/O. |
1da177e4 LT |
1504 | * |
1505 | */ | |
1506 | void submit_bio(int rw, struct bio *bio) | |
1507 | { | |
1508 | int count = bio_sectors(bio); | |
1509 | ||
22e2c507 | 1510 | bio->bi_rw |= rw; |
1da177e4 | 1511 | |
bf2de6f5 JA |
1512 | /* |
1513 | * If it's a regular read/write or a barrier with data attached, | |
1514 | * go through the normal accounting stuff before submission. | |
1515 | */ | |
a9c701e5 | 1516 | if (bio_has_data(bio)) { |
bf2de6f5 JA |
1517 | if (rw & WRITE) { |
1518 | count_vm_events(PGPGOUT, count); | |
1519 | } else { | |
1520 | task_io_account_read(bio->bi_size); | |
1521 | count_vm_events(PGPGIN, count); | |
1522 | } | |
1523 | ||
1524 | if (unlikely(block_dump)) { | |
1525 | char b[BDEVNAME_SIZE]; | |
1526 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | |
ba25f9dc | 1527 | current->comm, task_pid_nr(current), |
bf2de6f5 JA |
1528 | (rw & WRITE) ? "WRITE" : "READ", |
1529 | (unsigned long long)bio->bi_sector, | |
6728cb0e | 1530 | bdevname(bio->bi_bdev, b)); |
bf2de6f5 | 1531 | } |
1da177e4 LT |
1532 | } |
1533 | ||
1534 | generic_make_request(bio); | |
1535 | } | |
1da177e4 LT |
1536 | EXPORT_SYMBOL(submit_bio); |
1537 | ||
82124d60 KU |
1538 | /** |
1539 | * blk_rq_check_limits - Helper function to check a request for the queue limit | |
1540 | * @q: the queue | |
1541 | * @rq: the request being checked | |
1542 | * | |
1543 | * Description: | |
1544 | * @rq may have been made based on weaker limitations of upper-level queues | |
1545 | * in request stacking drivers, and it may violate the limitation of @q. | |
1546 | * Since the block layer and the underlying device driver trust @rq | |
1547 | * after it is inserted to @q, it should be checked against @q before | |
1548 | * the insertion using this generic function. | |
1549 | * | |
1550 | * This function should also be useful for request stacking drivers | |
1551 | * in some cases below, so export this fuction. | |
1552 | * Request stacking drivers like request-based dm may change the queue | |
1553 | * limits while requests are in the queue (e.g. dm's table swapping). | |
1554 | * Such request stacking drivers should check those requests agaist | |
1555 | * the new queue limits again when they dispatch those requests, | |
1556 | * although such checkings are also done against the old queue limits | |
1557 | * when submitting requests. | |
1558 | */ | |
1559 | int blk_rq_check_limits(struct request_queue *q, struct request *rq) | |
1560 | { | |
1561 | if (rq->nr_sectors > q->max_sectors || | |
1562 | rq->data_len > q->max_hw_sectors << 9) { | |
1563 | printk(KERN_ERR "%s: over max size limit.\n", __func__); | |
1564 | return -EIO; | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * queue's settings related to segment counting like q->bounce_pfn | |
1569 | * may differ from that of other stacking queues. | |
1570 | * Recalculate it to check the request correctly on this queue's | |
1571 | * limitation. | |
1572 | */ | |
1573 | blk_recalc_rq_segments(rq); | |
1574 | if (rq->nr_phys_segments > q->max_phys_segments || | |
1575 | rq->nr_phys_segments > q->max_hw_segments) { | |
1576 | printk(KERN_ERR "%s: over max segments limit.\n", __func__); | |
1577 | return -EIO; | |
1578 | } | |
1579 | ||
1580 | return 0; | |
1581 | } | |
1582 | EXPORT_SYMBOL_GPL(blk_rq_check_limits); | |
1583 | ||
1584 | /** | |
1585 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request | |
1586 | * @q: the queue to submit the request | |
1587 | * @rq: the request being queued | |
1588 | */ | |
1589 | int blk_insert_cloned_request(struct request_queue *q, struct request *rq) | |
1590 | { | |
1591 | unsigned long flags; | |
1592 | ||
1593 | if (blk_rq_check_limits(q, rq)) | |
1594 | return -EIO; | |
1595 | ||
1596 | #ifdef CONFIG_FAIL_MAKE_REQUEST | |
1597 | if (rq->rq_disk && rq->rq_disk->part0.make_it_fail && | |
1598 | should_fail(&fail_make_request, blk_rq_bytes(rq))) | |
1599 | return -EIO; | |
1600 | #endif | |
1601 | ||
1602 | spin_lock_irqsave(q->queue_lock, flags); | |
1603 | ||
1604 | /* | |
1605 | * Submitting request must be dequeued before calling this function | |
1606 | * because it will be linked to another request_queue | |
1607 | */ | |
1608 | BUG_ON(blk_queued_rq(rq)); | |
1609 | ||
1610 | drive_stat_acct(rq, 1); | |
1611 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | |
1612 | ||
1613 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1614 | ||
1615 | return 0; | |
1616 | } | |
1617 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); | |
1618 | ||
53a08807 TH |
1619 | /** |
1620 | * blkdev_dequeue_request - dequeue request and start timeout timer | |
1621 | * @req: request to dequeue | |
1622 | * | |
1623 | * Dequeue @req and start timeout timer on it. This hands off the | |
1624 | * request to the driver. | |
1625 | * | |
1626 | * Block internal functions which don't want to start timer should | |
1627 | * call elv_dequeue_request(). | |
1628 | */ | |
1629 | void blkdev_dequeue_request(struct request *req) | |
1630 | { | |
1631 | elv_dequeue_request(req->q, req); | |
1632 | ||
1633 | /* | |
1634 | * We are now handing the request to the hardware, add the | |
1635 | * timeout handler. | |
1636 | */ | |
1637 | blk_add_timer(req); | |
1638 | } | |
1639 | EXPORT_SYMBOL(blkdev_dequeue_request); | |
1640 | ||
bc58ba94 JA |
1641 | static void blk_account_io_completion(struct request *req, unsigned int bytes) |
1642 | { | |
26308eab | 1643 | if (!blk_do_io_stat(req)) |
bc58ba94 JA |
1644 | return; |
1645 | ||
1646 | if (blk_fs_request(req)) { | |
1647 | const int rw = rq_data_dir(req); | |
1648 | struct hd_struct *part; | |
1649 | int cpu; | |
1650 | ||
1651 | cpu = part_stat_lock(); | |
1652 | part = disk_map_sector_rcu(req->rq_disk, req->sector); | |
1653 | part_stat_add(cpu, part, sectors[rw], bytes >> 9); | |
1654 | part_stat_unlock(); | |
1655 | } | |
1656 | } | |
1657 | ||
1658 | static void blk_account_io_done(struct request *req) | |
1659 | { | |
26308eab | 1660 | if (!blk_do_io_stat(req)) |
bc58ba94 JA |
1661 | return; |
1662 | ||
1663 | /* | |
1664 | * Account IO completion. bar_rq isn't accounted as a normal | |
1665 | * IO on queueing nor completion. Accounting the containing | |
1666 | * request is enough. | |
1667 | */ | |
1668 | if (blk_fs_request(req) && req != &req->q->bar_rq) { | |
1669 | unsigned long duration = jiffies - req->start_time; | |
1670 | const int rw = rq_data_dir(req); | |
1671 | struct hd_struct *part; | |
1672 | int cpu; | |
1673 | ||
1674 | cpu = part_stat_lock(); | |
26308eab | 1675 | part = disk_map_sector_rcu(req->rq_disk, req->sector); |
bc58ba94 JA |
1676 | |
1677 | part_stat_inc(cpu, part, ios[rw]); | |
1678 | part_stat_add(cpu, part, ticks[rw], duration); | |
1679 | part_round_stats(cpu, part); | |
1680 | part_dec_in_flight(part); | |
1681 | ||
1682 | part_stat_unlock(); | |
1683 | } | |
1684 | } | |
1685 | ||
3bcddeac KU |
1686 | /** |
1687 | * __end_that_request_first - end I/O on a request | |
1688 | * @req: the request being processed | |
710027a4 | 1689 | * @error: %0 for success, < %0 for error |
3bcddeac KU |
1690 | * @nr_bytes: number of bytes to complete |
1691 | * | |
1692 | * Description: | |
1693 | * Ends I/O on a number of bytes attached to @req, and sets it up | |
1694 | * for the next range of segments (if any) in the cluster. | |
1695 | * | |
1696 | * Return: | |
710027a4 RD |
1697 | * %0 - we are done with this request, call end_that_request_last() |
1698 | * %1 - still buffers pending for this request | |
3bcddeac | 1699 | **/ |
5450d3e1 | 1700 | static int __end_that_request_first(struct request *req, int error, |
1da177e4 LT |
1701 | int nr_bytes) |
1702 | { | |
5450d3e1 | 1703 | int total_bytes, bio_nbytes, next_idx = 0; |
1da177e4 LT |
1704 | struct bio *bio; |
1705 | ||
5f3ea37c | 1706 | trace_block_rq_complete(req->q, req); |
2056a782 | 1707 | |
1da177e4 | 1708 | /* |
924cec77 TH |
1709 | * For fs requests, rq is just carrier of independent bio's |
1710 | * and each partial completion should be handled separately. | |
1711 | * Reset per-request error on each partial completion. | |
1712 | * | |
1713 | * TODO: tj: This is too subtle. It would be better to let | |
1714 | * low level drivers do what they see fit. | |
1da177e4 | 1715 | */ |
924cec77 | 1716 | if (blk_fs_request(req)) |
1da177e4 LT |
1717 | req->errors = 0; |
1718 | ||
6728cb0e JA |
1719 | if (error && (blk_fs_request(req) && !(req->cmd_flags & REQ_QUIET))) { |
1720 | printk(KERN_ERR "end_request: I/O error, dev %s, sector %llu\n", | |
1da177e4 LT |
1721 | req->rq_disk ? req->rq_disk->disk_name : "?", |
1722 | (unsigned long long)req->sector); | |
1723 | } | |
1724 | ||
bc58ba94 | 1725 | blk_account_io_completion(req, nr_bytes); |
d72d904a | 1726 | |
1da177e4 LT |
1727 | total_bytes = bio_nbytes = 0; |
1728 | while ((bio = req->bio) != NULL) { | |
1729 | int nbytes; | |
1730 | ||
1731 | if (nr_bytes >= bio->bi_size) { | |
1732 | req->bio = bio->bi_next; | |
1733 | nbytes = bio->bi_size; | |
5bb23a68 | 1734 | req_bio_endio(req, bio, nbytes, error); |
1da177e4 LT |
1735 | next_idx = 0; |
1736 | bio_nbytes = 0; | |
1737 | } else { | |
1738 | int idx = bio->bi_idx + next_idx; | |
1739 | ||
1740 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
1741 | blk_dump_rq_flags(req, "__end_that"); | |
6728cb0e | 1742 | printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", |
24c03d47 | 1743 | __func__, bio->bi_idx, bio->bi_vcnt); |
1da177e4 LT |
1744 | break; |
1745 | } | |
1746 | ||
1747 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | |
1748 | BIO_BUG_ON(nbytes > bio->bi_size); | |
1749 | ||
1750 | /* | |
1751 | * not a complete bvec done | |
1752 | */ | |
1753 | if (unlikely(nbytes > nr_bytes)) { | |
1754 | bio_nbytes += nr_bytes; | |
1755 | total_bytes += nr_bytes; | |
1756 | break; | |
1757 | } | |
1758 | ||
1759 | /* | |
1760 | * advance to the next vector | |
1761 | */ | |
1762 | next_idx++; | |
1763 | bio_nbytes += nbytes; | |
1764 | } | |
1765 | ||
1766 | total_bytes += nbytes; | |
1767 | nr_bytes -= nbytes; | |
1768 | ||
6728cb0e JA |
1769 | bio = req->bio; |
1770 | if (bio) { | |
1da177e4 LT |
1771 | /* |
1772 | * end more in this run, or just return 'not-done' | |
1773 | */ | |
1774 | if (unlikely(nr_bytes <= 0)) | |
1775 | break; | |
1776 | } | |
1777 | } | |
1778 | ||
1779 | /* | |
1780 | * completely done | |
1781 | */ | |
1782 | if (!req->bio) | |
1783 | return 0; | |
1784 | ||
1785 | /* | |
1786 | * if the request wasn't completed, update state | |
1787 | */ | |
1788 | if (bio_nbytes) { | |
5bb23a68 | 1789 | req_bio_endio(req, bio, bio_nbytes, error); |
1da177e4 LT |
1790 | bio->bi_idx += next_idx; |
1791 | bio_iovec(bio)->bv_offset += nr_bytes; | |
1792 | bio_iovec(bio)->bv_len -= nr_bytes; | |
1793 | } | |
1794 | ||
1795 | blk_recalc_rq_sectors(req, total_bytes >> 9); | |
1796 | blk_recalc_rq_segments(req); | |
1797 | return 1; | |
1798 | } | |
1799 | ||
1da177e4 LT |
1800 | /* |
1801 | * queue lock must be held | |
1802 | */ | |
5450d3e1 | 1803 | static void end_that_request_last(struct request *req, int error) |
1da177e4 | 1804 | { |
b8286239 KU |
1805 | if (blk_rq_tagged(req)) |
1806 | blk_queue_end_tag(req->q, req); | |
1807 | ||
1808 | if (blk_queued_rq(req)) | |
53a08807 | 1809 | elv_dequeue_request(req->q, req); |
1da177e4 LT |
1810 | |
1811 | if (unlikely(laptop_mode) && blk_fs_request(req)) | |
1812 | laptop_io_completion(); | |
1813 | ||
e78042e5 MA |
1814 | blk_delete_timer(req); |
1815 | ||
bc58ba94 | 1816 | blk_account_io_done(req); |
b8286239 | 1817 | |
1da177e4 | 1818 | if (req->end_io) |
8ffdc655 | 1819 | req->end_io(req, error); |
b8286239 KU |
1820 | else { |
1821 | if (blk_bidi_rq(req)) | |
1822 | __blk_put_request(req->next_rq->q, req->next_rq); | |
1823 | ||
1da177e4 | 1824 | __blk_put_request(req->q, req); |
b8286239 | 1825 | } |
1da177e4 LT |
1826 | } |
1827 | ||
3b11313a KU |
1828 | /** |
1829 | * blk_rq_bytes - Returns bytes left to complete in the entire request | |
5d87a052 | 1830 | * @rq: the request being processed |
3b11313a KU |
1831 | **/ |
1832 | unsigned int blk_rq_bytes(struct request *rq) | |
a0cd1285 JA |
1833 | { |
1834 | if (blk_fs_request(rq)) | |
1835 | return rq->hard_nr_sectors << 9; | |
1836 | ||
1837 | return rq->data_len; | |
1838 | } | |
3b11313a KU |
1839 | EXPORT_SYMBOL_GPL(blk_rq_bytes); |
1840 | ||
1841 | /** | |
1842 | * blk_rq_cur_bytes - Returns bytes left to complete in the current segment | |
5d87a052 | 1843 | * @rq: the request being processed |
3b11313a KU |
1844 | **/ |
1845 | unsigned int blk_rq_cur_bytes(struct request *rq) | |
1846 | { | |
1847 | if (blk_fs_request(rq)) | |
1848 | return rq->current_nr_sectors << 9; | |
1849 | ||
1850 | if (rq->bio) | |
1851 | return rq->bio->bi_size; | |
1852 | ||
1853 | return rq->data_len; | |
1854 | } | |
1855 | EXPORT_SYMBOL_GPL(blk_rq_cur_bytes); | |
a0cd1285 | 1856 | |
a0cd1285 JA |
1857 | /** |
1858 | * end_request - end I/O on the current segment of the request | |
8f731f7d | 1859 | * @req: the request being processed |
710027a4 | 1860 | * @uptodate: error value or %0/%1 uptodate flag |
a0cd1285 JA |
1861 | * |
1862 | * Description: | |
1863 | * Ends I/O on the current segment of a request. If that is the only | |
1864 | * remaining segment, the request is also completed and freed. | |
1865 | * | |
710027a4 RD |
1866 | * This is a remnant of how older block drivers handled I/O completions. |
1867 | * Modern drivers typically end I/O on the full request in one go, unless | |
a0cd1285 JA |
1868 | * they have a residual value to account for. For that case this function |
1869 | * isn't really useful, unless the residual just happens to be the | |
1870 | * full current segment. In other words, don't use this function in new | |
d00e29fd | 1871 | * code. Use blk_end_request() or __blk_end_request() to end a request. |
a0cd1285 JA |
1872 | **/ |
1873 | void end_request(struct request *req, int uptodate) | |
1874 | { | |
d00e29fd KU |
1875 | int error = 0; |
1876 | ||
1877 | if (uptodate <= 0) | |
1878 | error = uptodate ? uptodate : -EIO; | |
1879 | ||
1880 | __blk_end_request(req, error, req->hard_cur_sectors << 9); | |
a0cd1285 | 1881 | } |
1da177e4 LT |
1882 | EXPORT_SYMBOL(end_request); |
1883 | ||
32fab448 KU |
1884 | static int end_that_request_data(struct request *rq, int error, |
1885 | unsigned int nr_bytes, unsigned int bidi_bytes) | |
1886 | { | |
1887 | if (rq->bio) { | |
1888 | if (__end_that_request_first(rq, error, nr_bytes)) | |
1889 | return 1; | |
1890 | ||
1891 | /* Bidi request must be completed as a whole */ | |
1892 | if (blk_bidi_rq(rq) && | |
1893 | __end_that_request_first(rq->next_rq, error, bidi_bytes)) | |
1894 | return 1; | |
1895 | } | |
1896 | ||
1897 | return 0; | |
1898 | } | |
1899 | ||
336cdb40 | 1900 | /** |
e19a3ab0 KU |
1901 | * blk_end_io - Generic end_io function to complete a request. |
1902 | * @rq: the request being processed | |
710027a4 | 1903 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1904 | * @nr_bytes: number of bytes to complete @rq |
1905 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
e19a3ab0 KU |
1906 | * @drv_callback: function called between completion of bios in the request |
1907 | * and completion of the request. | |
710027a4 | 1908 | * If the callback returns non %0, this helper returns without |
e19a3ab0 | 1909 | * completion of the request. |
336cdb40 KU |
1910 | * |
1911 | * Description: | |
e3a04fe3 | 1912 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. |
336cdb40 KU |
1913 | * If @rq has leftover, sets it up for the next range of segments. |
1914 | * | |
1915 | * Return: | |
710027a4 RD |
1916 | * %0 - we are done with this request |
1917 | * %1 - this request is not freed yet, it still has pending buffers. | |
336cdb40 | 1918 | **/ |
22b13210 JA |
1919 | static int blk_end_io(struct request *rq, int error, unsigned int nr_bytes, |
1920 | unsigned int bidi_bytes, | |
1921 | int (drv_callback)(struct request *)) | |
336cdb40 KU |
1922 | { |
1923 | struct request_queue *q = rq->q; | |
1924 | unsigned long flags = 0UL; | |
336cdb40 | 1925 | |
32fab448 KU |
1926 | if (end_that_request_data(rq, error, nr_bytes, bidi_bytes)) |
1927 | return 1; | |
336cdb40 | 1928 | |
e19a3ab0 KU |
1929 | /* Special feature for tricky drivers */ |
1930 | if (drv_callback && drv_callback(rq)) | |
1931 | return 1; | |
1932 | ||
336cdb40 KU |
1933 | add_disk_randomness(rq->rq_disk); |
1934 | ||
1935 | spin_lock_irqsave(q->queue_lock, flags); | |
b8286239 | 1936 | end_that_request_last(rq, error); |
336cdb40 KU |
1937 | spin_unlock_irqrestore(q->queue_lock, flags); |
1938 | ||
1939 | return 0; | |
1940 | } | |
e19a3ab0 KU |
1941 | |
1942 | /** | |
1943 | * blk_end_request - Helper function for drivers to complete the request. | |
1944 | * @rq: the request being processed | |
710027a4 | 1945 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
1946 | * @nr_bytes: number of bytes to complete |
1947 | * | |
1948 | * Description: | |
1949 | * Ends I/O on a number of bytes attached to @rq. | |
1950 | * If @rq has leftover, sets it up for the next range of segments. | |
1951 | * | |
1952 | * Return: | |
710027a4 RD |
1953 | * %0 - we are done with this request |
1954 | * %1 - still buffers pending for this request | |
e19a3ab0 | 1955 | **/ |
22b13210 | 1956 | int blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
e19a3ab0 | 1957 | { |
e3a04fe3 | 1958 | return blk_end_io(rq, error, nr_bytes, 0, NULL); |
e19a3ab0 | 1959 | } |
336cdb40 KU |
1960 | EXPORT_SYMBOL_GPL(blk_end_request); |
1961 | ||
1962 | /** | |
1963 | * __blk_end_request - Helper function for drivers to complete the request. | |
1964 | * @rq: the request being processed | |
710027a4 | 1965 | * @error: %0 for success, < %0 for error |
336cdb40 KU |
1966 | * @nr_bytes: number of bytes to complete |
1967 | * | |
1968 | * Description: | |
1969 | * Must be called with queue lock held unlike blk_end_request(). | |
1970 | * | |
1971 | * Return: | |
710027a4 RD |
1972 | * %0 - we are done with this request |
1973 | * %1 - still buffers pending for this request | |
336cdb40 | 1974 | **/ |
22b13210 | 1975 | int __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) |
336cdb40 | 1976 | { |
60540161 | 1977 | if (rq->bio && __end_that_request_first(rq, error, nr_bytes)) |
051cc395 | 1978 | return 1; |
336cdb40 KU |
1979 | |
1980 | add_disk_randomness(rq->rq_disk); | |
1981 | ||
b8286239 | 1982 | end_that_request_last(rq, error); |
336cdb40 KU |
1983 | |
1984 | return 0; | |
1985 | } | |
1986 | EXPORT_SYMBOL_GPL(__blk_end_request); | |
1987 | ||
e3a04fe3 KU |
1988 | /** |
1989 | * blk_end_bidi_request - Helper function for drivers to complete bidi request. | |
1990 | * @rq: the bidi request being processed | |
710027a4 | 1991 | * @error: %0 for success, < %0 for error |
e3a04fe3 KU |
1992 | * @nr_bytes: number of bytes to complete @rq |
1993 | * @bidi_bytes: number of bytes to complete @rq->next_rq | |
1994 | * | |
1995 | * Description: | |
1996 | * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. | |
1997 | * | |
1998 | * Return: | |
710027a4 RD |
1999 | * %0 - we are done with this request |
2000 | * %1 - still buffers pending for this request | |
e3a04fe3 | 2001 | **/ |
22b13210 JA |
2002 | int blk_end_bidi_request(struct request *rq, int error, unsigned int nr_bytes, |
2003 | unsigned int bidi_bytes) | |
e3a04fe3 KU |
2004 | { |
2005 | return blk_end_io(rq, error, nr_bytes, bidi_bytes, NULL); | |
2006 | } | |
2007 | EXPORT_SYMBOL_GPL(blk_end_bidi_request); | |
2008 | ||
32fab448 KU |
2009 | /** |
2010 | * blk_update_request - Special helper function for request stacking drivers | |
2011 | * @rq: the request being processed | |
2012 | * @error: %0 for success, < %0 for error | |
2013 | * @nr_bytes: number of bytes to complete @rq | |
2014 | * | |
2015 | * Description: | |
2016 | * Ends I/O on a number of bytes attached to @rq, but doesn't complete | |
2017 | * the request structure even if @rq doesn't have leftover. | |
2018 | * If @rq has leftover, sets it up for the next range of segments. | |
2019 | * | |
2020 | * This special helper function is only for request stacking drivers | |
2021 | * (e.g. request-based dm) so that they can handle partial completion. | |
2022 | * Actual device drivers should use blk_end_request instead. | |
2023 | */ | |
2024 | void blk_update_request(struct request *rq, int error, unsigned int nr_bytes) | |
2025 | { | |
2026 | if (!end_that_request_data(rq, error, nr_bytes, 0)) { | |
2027 | /* | |
2028 | * These members are not updated in end_that_request_data() | |
2029 | * when all bios are completed. | |
2030 | * Update them so that the request stacking driver can find | |
2031 | * how many bytes remain in the request later. | |
2032 | */ | |
2033 | rq->nr_sectors = rq->hard_nr_sectors = 0; | |
2034 | rq->current_nr_sectors = rq->hard_cur_sectors = 0; | |
2035 | } | |
2036 | } | |
2037 | EXPORT_SYMBOL_GPL(blk_update_request); | |
2038 | ||
e19a3ab0 KU |
2039 | /** |
2040 | * blk_end_request_callback - Special helper function for tricky drivers | |
2041 | * @rq: the request being processed | |
710027a4 | 2042 | * @error: %0 for success, < %0 for error |
e19a3ab0 KU |
2043 | * @nr_bytes: number of bytes to complete |
2044 | * @drv_callback: function called between completion of bios in the request | |
2045 | * and completion of the request. | |
710027a4 | 2046 | * If the callback returns non %0, this helper returns without |
e19a3ab0 KU |
2047 | * completion of the request. |
2048 | * | |
2049 | * Description: | |
2050 | * Ends I/O on a number of bytes attached to @rq. | |
2051 | * If @rq has leftover, sets it up for the next range of segments. | |
2052 | * | |
2053 | * This special helper function is used only for existing tricky drivers. | |
2054 | * (e.g. cdrom_newpc_intr() of ide-cd) | |
2055 | * This interface will be removed when such drivers are rewritten. | |
2056 | * Don't use this interface in other places anymore. | |
2057 | * | |
2058 | * Return: | |
710027a4 RD |
2059 | * %0 - we are done with this request |
2060 | * %1 - this request is not freed yet. | |
2061 | * this request still has pending buffers or | |
2062 | * the driver doesn't want to finish this request yet. | |
e19a3ab0 | 2063 | **/ |
22b13210 JA |
2064 | int blk_end_request_callback(struct request *rq, int error, |
2065 | unsigned int nr_bytes, | |
e19a3ab0 KU |
2066 | int (drv_callback)(struct request *)) |
2067 | { | |
e3a04fe3 | 2068 | return blk_end_io(rq, error, nr_bytes, 0, drv_callback); |
e19a3ab0 KU |
2069 | } |
2070 | EXPORT_SYMBOL_GPL(blk_end_request_callback); | |
2071 | ||
86db1e29 JA |
2072 | void blk_rq_bio_prep(struct request_queue *q, struct request *rq, |
2073 | struct bio *bio) | |
1da177e4 | 2074 | { |
d628eaef DW |
2075 | /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw, and |
2076 | we want BIO_RW_AHEAD (bit 1) to imply REQ_FAILFAST (bit 1). */ | |
4aff5e23 | 2077 | rq->cmd_flags |= (bio->bi_rw & 3); |
1da177e4 | 2078 | |
fb2dce86 DW |
2079 | if (bio_has_data(bio)) { |
2080 | rq->nr_phys_segments = bio_phys_segments(q, bio); | |
fb2dce86 DW |
2081 | rq->buffer = bio_data(bio); |
2082 | } | |
1da177e4 LT |
2083 | rq->current_nr_sectors = bio_cur_sectors(bio); |
2084 | rq->hard_cur_sectors = rq->current_nr_sectors; | |
2085 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | |
0e75f906 | 2086 | rq->data_len = bio->bi_size; |
1da177e4 LT |
2087 | |
2088 | rq->bio = rq->biotail = bio; | |
1da177e4 | 2089 | |
66846572 N |
2090 | if (bio->bi_bdev) |
2091 | rq->rq_disk = bio->bi_bdev->bd_disk; | |
2092 | } | |
1da177e4 | 2093 | |
ef9e3fac KU |
2094 | /** |
2095 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy | |
2096 | * @q : the queue of the device being checked | |
2097 | * | |
2098 | * Description: | |
2099 | * Check if underlying low-level drivers of a device are busy. | |
2100 | * If the drivers want to export their busy state, they must set own | |
2101 | * exporting function using blk_queue_lld_busy() first. | |
2102 | * | |
2103 | * Basically, this function is used only by request stacking drivers | |
2104 | * to stop dispatching requests to underlying devices when underlying | |
2105 | * devices are busy. This behavior helps more I/O merging on the queue | |
2106 | * of the request stacking driver and prevents I/O throughput regression | |
2107 | * on burst I/O load. | |
2108 | * | |
2109 | * Return: | |
2110 | * 0 - Not busy (The request stacking driver should dispatch request) | |
2111 | * 1 - Busy (The request stacking driver should stop dispatching request) | |
2112 | */ | |
2113 | int blk_lld_busy(struct request_queue *q) | |
2114 | { | |
2115 | if (q->lld_busy_fn) | |
2116 | return q->lld_busy_fn(q); | |
2117 | ||
2118 | return 0; | |
2119 | } | |
2120 | EXPORT_SYMBOL_GPL(blk_lld_busy); | |
2121 | ||
18887ad9 | 2122 | int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) |
1da177e4 LT |
2123 | { |
2124 | return queue_work(kblockd_workqueue, work); | |
2125 | } | |
1da177e4 LT |
2126 | EXPORT_SYMBOL(kblockd_schedule_work); |
2127 | ||
1da177e4 LT |
2128 | int __init blk_dev_init(void) |
2129 | { | |
2130 | kblockd_workqueue = create_workqueue("kblockd"); | |
2131 | if (!kblockd_workqueue) | |
2132 | panic("Failed to create kblockd\n"); | |
2133 | ||
2134 | request_cachep = kmem_cache_create("blkdev_requests", | |
20c2df83 | 2135 | sizeof(struct request), 0, SLAB_PANIC, NULL); |
1da177e4 | 2136 | |
8324aa91 | 2137 | blk_requestq_cachep = kmem_cache_create("blkdev_queue", |
165125e1 | 2138 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
1da177e4 | 2139 | |
d38ecf93 | 2140 | return 0; |
1da177e4 | 2141 | } |
1da177e4 | 2142 |